AI-powered environmental and health intelligence built using Google’s ADK.
Over 150 million Americans live in regions with poor air quality — yet daily health decisions still rely on fragmented tools:
- Weather apps show temperature but not health implications
- AQI apps show pollution but not symptoms
- Health sites offer generic advice without environmental context
- Users often check 4–5 different apps daily
- Provide raw data, not intelligence
- No correlation between environment → symptoms → outbreaks
- No medical triage contextualized to environmental triggers
EcoGuardian is an AI-driven, multi-agent environmental health intelligence platform built with Google ADK.
- 30× faster than manually checking multiple apps
- Unified analysis across air, weather, pollen, UV, symptoms, outbreaks
- Smart routing → only the required agents are triggered
- Real-time outbreak intelligence (WHO, CDC, GDELT)
- Safety-first medical triage with mandatory disclaimers
- Routes queries to proper agents
- Maintains
{city: str}in session state - Synthesizes multi-agent responses
- Air Quality Agent (OpenAQ)
- Weather Agent (Open-Meteo)
- Pollen Agent (Pollen.com)
- UV Index Agent (Open-Meteo AQ)
Stage 1 – Parallel Execution
- Symptom Analyzer
- Outbreak Monitor
Stage 2 – Sequential
- Hospital Locator
Stage 3 – Decision Coordinator
- Merges intel into actionable guidance
- Uses google_search tool for environmental/sustainability events
https://ecoguardian-mqvs64cf3q-uc.a.run.app/
- Real-time AQI + PM2.5, PM10, O₃, NO₂
- Temperature, humidity, wind, precipitation
- Pollen levels (tree/grass/weed)
- UV index + risk analysis
- Disease outbreak detection (WHO, CDC, GDELT)
- Symptom-based urgency classification (NOT diagnosis)
- Nearest hospitals with distance + emergency flagging
- Built-in medical safety guardrails
- Streamlit UI
- Persistent session storage (SQLite)
- Conversation memory
- Multi-agent orchestration (10 agents)
git clone https://github.com/satvika-eda/EcoGuardian.git
cd EcoGuardianpython -m venv .venv
source .venv/bin/activate pip install -r requirements.txtAdd GOOGLE_API_KEY and OPENAQ_API_KEY
GOOGLE_API_KEY=your_key
OPENAQ_API_KEY=your_key
- Google Gemini API
- OpenAQ Air Quality API
- Open-Meteo Weather & UV API
- Pollen.com API (unofficial)
- WHO + CDC + GDELT outbreak data
- OpenStreetMap Overpass API
streamlit run app.py- "How's the environment in Miami today?"
- "I have fever and cough for 8 days in Miami."
- "Find nearest hospitals."
- "What sustainability events are happening this weekend?"
| Component | Technology |
|---|---|
| Framework | Google ADK |
| LLM | Gemini 2.5 Flash Lite |
| UI | Streamlit |
| Database | SQLite |
| Memory | InMemoryMemoryService |
| Orchestration | LlmAgent, SequentialAgent, ParallelAgent |
| APIs | OpenAQ, Open-Meteo, Pollen.com, WHO, CDC, GDELT, OSM |
- Independent agents
- Sequential pipelines
- Parallel execution
- Hybrid routing
- Agents wrapped as tools (
AgentTool)
Each API is accessed via an ADK Tool() or AgentTool().
- Multi-Agent System (10 agents total)
- SequentialAgent (3-stage medical pipeline)
- ParallelAgent (symptoms + outbreaks)
- AgentTool wrappers
- Session persistence (SQLite via DatabaseSessionService)
- Memory (InMemoryMemoryService)
- State management (StateSchema: {city: str})
- Observability (LoggingPlugin)
- Retry logic for API failures
- Medical safety guardrails
- Evaluation Suite
- Agent Deployment
- Diagnoses medical conditions
- Recommends treatments
- Provides prescriptions
- Replaces doctors
- Adds medical disclaimers
- Flags emergencies
- Encourages professional care
- Avoids PHI storage
- Uses anonymous session IDs only
- Integrate Google Maps to the UI for better user experience.
- Add more agents for other environmental and health factors
- Caching
- A2A protocol (Agent-to-Agent)
- Monitoring dashboard (Grafana + Prometheus)